Luepol Pipanmaekaporn

Feature discovery in relevance feedback using pattern mining

It is a big challenge to guarantee the quality of extracted features in text documents to describe user interests or preferences due to large amounts of noise. Over the years, pattern mining-based approaches to RF have attracted great interest to discover knowledge of user interest from text documents. However, the data mining approaches often produce a large set of patterns, which include a lot of noisy patterns, reducing the effective use of pattern mining. In this paper, we present a novel pattern mining approach to RF. This approach mines patterns in both positive and negative feedback and then classifies them into clusters to find user-specific patterns. We also propose a novel pattern deploying method that effectively uses the discovered patterns for improving the performance of searching relevant documents. Experiments are conducted on Reuters Corpus Volume 1 data collection (RCV1) and TREC filtering topics. The results show that the proposed approach achieves promising performance comparing with state-of-the-art term-based methods and pattern-based ones.

Generating Test Cases from UML Activity Diagram Based on Business Flow Constraints

Test case generation is the most important part of software testing. More than 50 percent of the cost and time are spent on testing the software development. Currently, researchers have used the UML activity for test case generation. However, finding a test case set from an activity diagram is a terrible task. Because the presence of loop and concurrent activities in the activity diagram result in path explosion and practically, it is not feasible to consider all execution paths for testing. In this paper, we proposed a novel approach to generate test cases using a modified DFS with tester specification to avoid the path explosion. Tester specifications follow the business flow constraints. In order to evaluate the quality of test cases, activity coverage, transition coverage, and key path coverage are measured. The proposed approach shows that the result paths of an activity diagram cover both true and false values of loop condition and helps to avoid generating all possible concurrent activity paths as it only returns one representative path.

Learning high-level features for chord recognition using Autoencoder

Chord transcription is valuable to do by itself. It is known that the manual transcription of chords is very tiresome, time-consuming. It requires, moreover, musical knowledge. Automatic chord recognition has recently attracted a number of researches in the Music Information Retrieval field. It has known that a pitch class profile (PCP) is the commonly signal representation of musical harmonic analysis. However, the PCP may contain additional non-harmonic noise such as harmonic overtones and transient noise. The problem of non-harmonic might be generating the sound energy in term of frequency more than the actual notes of the respective chord. Autoencoder neural network may be trained to learn a mapping from low level feature to one or more higher-level representation. These high-level representations can explain dependencies of the inputs and reduce the effect of non-harmonic noise. Then these improve features are fed into neural network classifier. The proposed high-level musical features show 80.90% of accuracy. The experimental results have shown that the proposed approach can achieve better performance in comparison with other based method.

Classification of Drowsiness in EEG Records Based on Energy Distribution and Wavelet-Neural Network

Drowsiness is the main factors in traffic accidents because the ability of vehicle driver was diminished. These conditions will endanger to own driver and the other vehicle drivers. With the growing traffic conditions this problem will increase in the future. So, it is important to develop automatic characterization of the drowsiness stage. The aim of this paper presents a new method to improve wavelet coefficient of DWT for classification alert and drowsiness stages of EEG signals. The method applied the Parsevals theorem and energy coefficient distribution. The Input-Output cluster method was used to estimate the approximate status of each input features. Then these improve features are feeded into neural network classifier. The proposed method gets 90.27% of accuracy. The experimental results have shown that the proposed approach can achieve better performance in comparison with other based methods.

Deploying Top-k Specific Patterns for Relevance Feature Discovery

The quality of discovered patterns is important for relevance feature discovery in text documents because frequent pattern mining often produces many noisy patterns. In this paper, we propose a novel method for the summarization of discovered patterns in text documents, finding a smaller number of specific patterns for representing the large number of discovered patterns for a given topic. We also evaluate the proposed method by implementing a new pattern-based information filtering model. The experimental results show that the proposed method not only outperforms both the term-based approaches and pattern based approaches, but largely reduces the number of feature terms as well.

Concurrent test case generation from UML activity diagram based on dynamic programming

Concurrency is a challenging and difficult task for software testing. Many researchers try to solve this problem and propose a solution, i.e. control flow graph, breadth first search (BFS), combined activity diagram and I/O explicit activity diagram. This paper proposes the method to generate concurrent test cases from UML activity diagrams. This method first builds a concurrent activity diagram from an ordinary UML activity diagram. It then transforms a UML activity diagram to an activity graph. Using dynamic programming techniques to find the number of total paths and of total condition paths in concurrent test cases. Test coverage measures the amount of tests performed by a set of tests which can also assure the quality of tests.

Time-frequency ratio hashing for content-based audio retrieval

Several approaches have previously been taken to the problem of content-based audio retrieval. These have used different features for indexing and searching. However, due to the nonstationarities and discontinuities exist in the audio signals, the retrieval strategy remains a formidable challenge. In this paper, we propose a novel fingerprint hash model with the concept of time-frequency ratio for audio matching in an attempt to achieve high retrieval accuracy. This model is examined using a database composed of variety of eight genres of 300 audio clips, collected from ballroom dataset. The results of the experiment support the effectiveness for audio retrieval with high precision and recall values at 0.99 and 0.95 respectively.

Improving Pitch Class Profile for Musical Chords Recognition Combining Major Chord Filters and Convolution Neural Networks

In this paper, we consider the challenging problem of music recognition and present an effective deep learning based method using a convolution neural network for chord recognition. It has known that a pitch class profile (PCP) is the commonly signal representation of musical harmonic analysis. However, the PCP vector is not expressive enough for chord recognition, which often occurs in many real-world environments. In this study, we extend the PCP vector scheme to address the limitation. Our proposed method basically consists of two major steps. First, we introduce novel filters and apply then to PCP vector to transform the vector into membership of 7 major chords as features to represent the input matrix. The second step is to efficiently learning feature on the transformed matrix (2D-PCP) using convolution neural network. We propose a trainable, data-driven approach that automatically learns features and its classifier simultaneously. Experimental results conducted on the task of musical chords recognition that the proposed method achieves improvements of classification accuracy more than 40% in accuracy in comparing with based line methods.

Trajectory mining from VMS data for identifying fishing tackles

Automatic identification of fishing equipment has a big impact on fisheries managements and illegal fishing surveillance. For many years, existing approaches to recognize fishing gear types have been proposed based on analysis of Vessel Monitoring System (VMS) data. However, the ship tracking data typically contain irrelevant and meaningless information that can limit their effectiveness. An innovative approach present in this paper is to identify types of fishing equipment from VMS records. Our approach first tries to identify activities of interest in a fishing using an unsupervised way. It then generates possible trajectories for the local movements and performs feature extraction. Two types of trajectory-based features are extracted to describe both global and local characteristics of fishing movement patterns. We finally perform dimension reduction and build the classifier using machine learning. Experiments conducted on historical VMS records from 180 commercial fishing boats with three major types of fishing gears in Thailand show that our approach achieves encouraging performance of recognition rates.

Learning Representation for fMRI Data Analysis Using Autoencoder

Analysis of fMRI data is very useful for studying relationship between neural activity and a variety of brain functions. For many years, a number of brain image analysis techniques using machine learning were proposed. However, this task is still challenging due to the unique characteristics of the brain data with very small samples but extremely high dimensionality, reducing generalization performance. This paper presents a novel analysis method for fMRI data. It consists of three major steps: (1) Identifying informative voxels, (2) extracting feature space by analyzing semantic relationships among voxels and (3) learning fMRI classifier from the extracted features. Preliminary experimental results conducted on the task of image prediction from fMRI data confirmed that the proposed method achieves improvements of classification accuracy more than 20% in mean accuracy in comparing with current neuroimaging methods.